Breast cancer histopathological image classification using attention high‐order deep network

• DOI: 10.1002/ima.22628
• Publication Date: 2021-07-16T06:00:21Z
• Authors: Ying Zou; Jianxin Zhang; Shan Huang; Bin Liu
• Keywords: Digital Pathology

Abstract Computer‐aided classification of pathological images is the great significance for breast cancer diagnosis. In recent years, deep learning methods image have made breakthrough progress, becoming mainstream in this field. To capture more discriminant features images, work introduces a novel attention high‐order network (AHoNet) by simultaneously embedding mechanism and statistical representation into residual convolutional network. AHoNet firstly employs an efficient channel module with non‐dimensionality reduction local cross‐channel interaction to achieve salient images. Then, their second‐order covariance statistics are further estimated through matrix power normalization, which provides robust global feature presentation We extensively evaluate on public BreakHis BACH pathology datasets. Experimental results illustrate that gains optimal patient‐level accuracies 99.29% 85% database, respectively, demonstrating competitive performance state‐of‐the‐art single models medical application.